通过域适应实现人脸识别

发布时间：2018-05-09 17:04

本文选题：人脸识别 + 域适应　；参考：《计算机应用研究》2017年06期

【摘要】：当源域和目的域数据分布不同时,大多数机器学习方法的性能会降低。为了解决这一问题,基于域适应的思想,提出了一种新的人脸识别方法。首先计算源域样本的相对权值,删除与目的域样本相差很大的样本,降低两域之间的差异性;然后采用基于正规化的Bregman divergence获得公共子空间,获得两域之间的共性;最后利用目的域样本目标化源域样本,充分利用目的域的特有信息。在此基础上建立的分类模型能够充分利用两域之间的共性和目的域的特性,实现对目的域的准确分类。为了评估方法的性能,在多个数据集上进行测试实验。实验结果证明,该方法的性能与其他几种方法相比均有所提高。
[Abstract]:When the data distribution between source domain and destination domain is different, the performance of most machine learning methods will decrease. In order to solve this problem, a new face recognition method is proposed based on the idea of domain adaptation. First, the relative weights of the source domain samples are calculated, the samples which are quite different from the target domain samples are deleted, the differences between the two domains are reduced, and then the common subspace is obtained by using the regularized Bregman divergence to obtain the commonality between the two domains. Finally, the target domain samples are used to target the source domain samples, and the specific information of the target domain is fully utilized. On this basis, the classification model can make full use of the commonness between the two domains and the characteristics of the target domain, and realize the accurate classification of the target domain. In order to evaluate the performance of the method, test experiments are carried out on multiple data sets. The experimental results show that the performance of this method is improved compared with other methods.
【作者单位】：中国矿业大学计算机科学技术学院;
【基金】：国家自然科学基金委员会—山西省人民政府煤基低碳联合基金资助项目(U1510115) 中国博士后科学基金特别资助项目(2013T60574)
【分类号】：TP391.41

【参考文献】